Glossary for Geospatial Science

nearest neighbor sampling/interpolation:  Resampling a source raster to yield a new raster with a different cell size, raster orientation, and/or internal geometry by computing the distance between the center of each cell in the output raster and the 4 nearest cells in the input raster.  The data value for the closest input cell is assigned without alteration to be the data value of the output cell.  Therefore, the input value of one input cell may be assigned to more than one output cell.  It also means that some input cells may not be transferred at all to the output raster.  These undersampling and oversampling situations occur when the cell sizes of the input and output raster are different.

For example, a resampled output raster which is coparallel to the input raster but which has cells half as big, will repeat the input values in a 2 x 2 cell pattern.  However, since nearest neighbor resampling does not mathematically compute the new cell’s value, it is the only suitable choice for rasters that are not mathematically continuous such as land cover maps.  With such maps, mathematical resampling (like bilinear, or cubic convolution) is totally incorrect since mean values cannot be computed.